4 Trivial Database 2: private types and prototypes
5 Copyright (C) Rusty Russell 2010
7 This library is free software; you can redistribute it and/or
8 modify it under the terms of the GNU Lesser General Public
9 License as published by the Free Software Foundation; either
10 version 3 of the License, or (at your option) any later version.
12 This library is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public
18 License along with this library; if not, see <http://www.gnu.org/licenses/>.
35 #include <ccan/tdb2/tdb2.h>
36 #include <ccan/likely/likely.h>
37 #include <ccan/compiler/compiler.h>
46 /* #define TDB_TRACE 1 */
49 #define __STRING(x) #x
52 #ifndef __STRINGSTRING
53 #define __STRINGSTRING(x) __STRING(x)
57 #define __location__ __FILE__ ":" __STRINGSTRING(__LINE__)
60 typedef uint64_t tdb_len_t;
61 typedef uint64_t tdb_off_t;
63 #define TDB_MAGIC_FOOD "TDB file\n"
64 #define TDB_VERSION ((uint64_t)(0x26011967 + 7))
65 #define TDB_USED_MAGIC ((uint64_t)0x1999)
66 #define TDB_HTABLE_MAGIC ((uint64_t)0x1888)
67 #define TDB_CHAIN_MAGIC ((uint64_t)0x1777)
68 #define TDB_FTABLE_MAGIC ((uint64_t)0x1666)
69 #define TDB_FREE_MAGIC ((uint64_t)0xFE)
70 #define TDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
71 #define TDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL)
72 #define TDB_RECOVERY_INVALID_MAGIC (0x0ULL)
74 #define TDB_OFF_IS_ERR(off) unlikely(off >= (tdb_off_t)TDB_ERR_LAST)
76 /* Packing errors into pointers and v.v. */
77 #define TDB_PTR_IS_ERR(ptr) \
78 unlikely((unsigned long)(ptr) >= (unsigned long)TDB_ERR_LAST)
79 #define TDB_PTR_ERR(p) ((enum TDB_ERROR)(long)(p))
80 #define TDB_ERR_PTR(err) ((void *)(long)(err))
82 /* Common case of returning true, false or -ve error. */
83 typedef int tdb_bool_err;
85 /* Prevent others from opening the file. */
86 #define TDB_OPEN_LOCK 0
87 /* Doing a transaction. */
88 #define TDB_TRANSACTION_LOCK 1
90 #define TDB_EXPANSION_LOCK 2
91 /* Hash chain locks. */
92 #define TDB_HASH_LOCK_START 3
94 /* Range for hash locks. */
95 #define TDB_HASH_LOCK_RANGE_BITS 30
96 #define TDB_HASH_LOCK_RANGE (1 << TDB_HASH_LOCK_RANGE_BITS)
98 /* We have 1024 entries in the top level. */
99 #define TDB_TOPLEVEL_HASH_BITS 10
100 /* And 64 entries in each sub-level: thus 64 bits exactly after 9 levels. */
101 #define TDB_SUBLEVEL_HASH_BITS 6
102 /* And 8 entries in each group, ie 8 groups per sublevel. */
103 #define TDB_HASH_GROUP_BITS 3
104 /* This is currently 10: beyond this we chain. */
105 #define TDB_MAX_LEVELS (1+(64-TDB_TOPLEVEL_HASH_BITS) / TDB_SUBLEVEL_HASH_BITS)
107 /* Extend file by least 100 times larger than needed. */
108 #define TDB_EXTENSION_FACTOR 100
110 /* We steal bits from the offsets to store hash info. */
111 #define TDB_OFF_HASH_GROUP_MASK ((1ULL << TDB_HASH_GROUP_BITS) - 1)
112 /* We steal this many upper bits, giving a maximum offset of 64 exabytes. */
113 #define TDB_OFF_UPPER_STEAL 8
114 #define TDB_OFF_UPPER_STEAL_EXTRA 7
115 /* The bit number where we store extra hash bits. */
116 #define TDB_OFF_HASH_EXTRA_BIT 57
117 #define TDB_OFF_UPPER_STEAL_SUBHASH_BIT 56
119 /* Additional features we understand. Currently: none. */
120 #define TDB_FEATURE_MASK ((uint64_t)0)
122 /* The bit number where we store the extra hash bits. */
123 /* Convenience mask to get actual offset. */
124 #define TDB_OFF_MASK \
125 (((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1) - TDB_OFF_HASH_GROUP_MASK)
127 /* How many buckets in a free list: see size_to_bucket(). */
128 #define TDB_FREE_BUCKETS (64 - TDB_OFF_UPPER_STEAL)
130 /* We have to be able to fit a free record here. */
131 #define TDB_MIN_DATA_LEN \
132 (sizeof(struct tdb_free_record) - sizeof(struct tdb_used_record))
134 /* Indicates this entry is not on an flist (can happen during coalescing) */
135 #define TDB_FTABLE_NONE ((1ULL << TDB_OFF_UPPER_STEAL) - 1)
138 static inline uint64_t bswap_64(uint64_t x)
140 return (((x&0x000000FFULL)<<56)
141 | ((x&0x0000FF00ULL)<<48)
142 | ((x&0x00FF0000ULL)<<40)
143 | ((x&0xFF000000ULL)<<32)
144 | ((x>>8)&0xFF000000ULL)
145 | ((x>>16)&0x00FF0000ULL)
146 | ((x>>24)&0x0000FF00ULL)
147 | ((x>>32)&0x000000FFULL));
151 struct tdb_used_record {
152 /* For on-disk compatibility, we avoid bitfields:
158 uint64_t magic_and_meta;
159 /* The bottom key_len_bits*2 are key length, rest is data length. */
160 uint64_t key_and_data_len;
163 static inline unsigned rec_key_bits(const struct tdb_used_record *r)
165 return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
168 static inline uint64_t rec_key_length(const struct tdb_used_record *r)
170 return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
173 static inline uint64_t rec_data_length(const struct tdb_used_record *r)
175 return r->key_and_data_len >> rec_key_bits(r);
178 static inline uint64_t rec_extra_padding(const struct tdb_used_record *r)
180 return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
183 static inline uint32_t rec_hash(const struct tdb_used_record *r)
185 return r->magic_and_meta & ((1 << 11) - 1);
188 static inline uint16_t rec_magic(const struct tdb_used_record *r)
190 return (r->magic_and_meta >> 48);
193 struct tdb_free_record {
194 uint64_t magic_and_prev; /* TDB_OFF_UPPER_STEAL bits magic, then prev */
195 uint64_t ftable_and_len; /* Len not counting these two fields. */
196 /* This is why the minimum record size is 8 bytes. */
200 static inline uint64_t frec_prev(const struct tdb_free_record *f)
202 return f->magic_and_prev & ((1ULL << (64 - TDB_OFF_UPPER_STEAL)) - 1);
205 static inline uint64_t frec_magic(const struct tdb_free_record *f)
207 return f->magic_and_prev >> (64 - TDB_OFF_UPPER_STEAL);
210 static inline uint64_t frec_len(const struct tdb_free_record *f)
212 return f->ftable_and_len & ((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
215 static inline unsigned frec_ftable(const struct tdb_free_record *f)
217 return f->ftable_and_len >> (64 - TDB_OFF_UPPER_STEAL);
220 struct tdb_recovery_record {
222 /* Length of record. */
226 /* Old length of file before transaction. */
230 /* If we bottom out of the subhashes, we chain. */
232 tdb_off_t rec[1 << TDB_HASH_GROUP_BITS];
236 /* this is stored at the front of every database */
238 char magic_food[64]; /* for /etc/magic */
239 /* FIXME: Make me 32 bit? */
240 uint64_t version; /* version of the code */
241 uint64_t hash_test; /* result of hashing HASH_MAGIC. */
242 uint64_t hash_seed; /* "random" seed written at creation time. */
243 tdb_off_t free_table; /* (First) free table. */
244 tdb_off_t recovery; /* Transaction recovery area. */
246 uint64_t features_used; /* Features all writers understand */
247 uint64_t features_offered; /* Features offered */
249 uint64_t seqnum; /* Sequence number for TDB_SEQNUM */
251 tdb_off_t reserved[23];
253 /* Top level hash table. */
254 tdb_off_t hashtable[1ULL << TDB_TOPLEVEL_HASH_BITS];
257 struct tdb_freetable {
258 struct tdb_used_record hdr;
260 tdb_off_t buckets[TDB_FREE_BUCKETS];
263 /* Information about a particular (locked) hash entry. */
265 /* Full hash value of entry. */
267 /* Start and length of lock acquired. */
268 tdb_off_t hlock_start;
269 tdb_len_t hlock_range;
270 /* Start of hash group. */
271 tdb_off_t group_start;
272 /* Bucket we belong in. */
273 unsigned int home_bucket;
274 /* Bucket we (or an empty space) were found in. */
275 unsigned int found_bucket;
276 /* How many bits of the hash are already used. */
277 unsigned int hash_used;
278 /* Current working group. */
279 tdb_off_t group[1 << TDB_HASH_GROUP_BITS];
282 struct traverse_info {
283 struct traverse_level {
285 /* We ignore groups here, and treat it as a big array. */
287 unsigned int total_buckets;
288 } levels[TDB_MAX_LEVELS + 1];
289 unsigned int num_levels;
290 unsigned int toplevel_group;
291 /* This makes delete-everything-inside-traverse work as expected. */
295 enum tdb_lock_flags {
296 /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
299 /* If set, don't log an error on failure. */
301 /* If set, don't check for recovery (used by recovery code). */
302 TDB_LOCK_NOCHECK = 4,
306 struct tdb_context *owner;
312 /* This is only needed for tdb_access_commit, but used everywhere to
314 struct tdb_access_hdr {
315 struct tdb_access_hdr *next;
322 /* Single list of all TDBs, to detect multiple opens. */
323 struct tdb_file *next;
325 /* How many are sharing us? */
328 /* Mmap (if any), or malloc (for TDB_INTERNAL). */
331 /* How much space has been mapped (<= current file size) */
334 /* The file descriptor (-1 for TDB_INTERNAL). */
337 /* Lock information */
338 struct tdb_lock allrecord_lock;
340 struct tdb_lock *lockrecs;
342 /* Identity of this file. */
348 /* Filename of the database. */
351 /* Are we accessing directly? (debugging check). */
354 /* Operating read-only? (Opened O_RDONLY, or in traverse_read) */
357 /* mmap read only? */
360 /* Error code for last tdb error. */
361 enum TDB_ERROR ecode;
363 /* the flags passed to tdb_open, for tdb_reopen. */
366 /* Logging function */
367 void (*logfn)(struct tdb_context *tdb,
368 enum tdb_log_level level,
370 const char *message);
374 uint64_t (*khash)(const void *key, size_t len, uint64_t seed, void *);
378 /* Set if we are in a transaction. */
379 struct tdb_transaction *transaction;
381 /* What free table are we using? */
382 tdb_off_t ftable_off;
385 /* IO methods: changes for transactions. */
386 const struct tdb_methods *methods;
388 struct tdb_attribute_stats *stats;
390 /* Direct access information */
391 struct tdb_access_hdr *access;
393 /* Last error we returned. */
394 enum TDB_ERROR last_error;
396 /* The actual file information */
397 struct tdb_file *file;
401 enum TDB_ERROR (*tread)(struct tdb_context *, tdb_off_t, void *,
403 enum TDB_ERROR (*twrite)(struct tdb_context *, tdb_off_t, const void *,
405 enum TDB_ERROR (*oob)(struct tdb_context *, tdb_off_t, bool);
406 enum TDB_ERROR (*expand_file)(struct tdb_context *, tdb_len_t);
407 void *(*direct)(struct tdb_context *, tdb_off_t, size_t, bool);
414 void tdb_hash_init(struct tdb_context *tdb);
416 tdb_bool_err first_in_hash(struct tdb_context *tdb,
417 struct traverse_info *tinfo,
418 TDB_DATA *kbuf, size_t *dlen);
420 tdb_bool_err next_in_hash(struct tdb_context *tdb,
421 struct traverse_info *tinfo,
422 TDB_DATA *kbuf, size_t *dlen);
424 /* Hash random memory. */
425 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len);
428 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off);
430 /* Find and lock a hash entry (or where it would be). */
431 tdb_off_t find_and_lock(struct tdb_context *tdb,
435 struct tdb_used_record *rec,
436 struct traverse_info *tinfo);
438 enum TDB_ERROR replace_in_hash(struct tdb_context *tdb,
442 enum TDB_ERROR add_to_hash(struct tdb_context *tdb, struct hash_info *h,
445 enum TDB_ERROR delete_from_hash(struct tdb_context *tdb, struct hash_info *h);
448 bool is_subhash(tdb_off_t val);
451 enum TDB_ERROR tdb_ftable_init(struct tdb_context *tdb);
453 /* check.c needs these to iterate through free lists. */
454 tdb_off_t first_ftable(struct tdb_context *tdb);
455 tdb_off_t next_ftable(struct tdb_context *tdb, tdb_off_t ftable);
457 /* This returns space or -ve error number. */
458 tdb_off_t alloc(struct tdb_context *tdb, size_t keylen, size_t datalen,
459 uint64_t hash, unsigned magic, bool growing);
461 /* Put this record in a free list. */
462 enum TDB_ERROR add_free_record(struct tdb_context *tdb,
463 tdb_off_t off, tdb_len_t len_with_header);
465 /* Set up header for a used/ftable/htable/chain record. */
466 enum TDB_ERROR set_header(struct tdb_context *tdb,
467 struct tdb_used_record *rec,
468 unsigned magic, uint64_t keylen, uint64_t datalen,
469 uint64_t actuallen, unsigned hashlow);
471 /* Used by tdb_check to verify. */
472 unsigned int size_to_bucket(tdb_len_t data_len);
473 tdb_off_t bucket_off(tdb_off_t ftable_off, unsigned bucket);
475 /* Used by tdb_summary */
476 tdb_off_t dead_space(struct tdb_context *tdb, tdb_off_t off);
479 /* Initialize tdb->methods. */
480 void tdb_io_init(struct tdb_context *tdb);
482 /* Convert endian of the buffer if required. */
483 void *tdb_convert(const struct tdb_context *tdb, void *buf, tdb_len_t size);
485 /* Unmap and try to map the tdb. */
486 void tdb_munmap(struct tdb_file *file);
487 void tdb_mmap(struct tdb_context *tdb);
489 /* Either alloc a copy, or give direct access. Release frees or noop. */
490 const void *tdb_access_read(struct tdb_context *tdb,
491 tdb_off_t off, tdb_len_t len, bool convert);
492 void *tdb_access_write(struct tdb_context *tdb,
493 tdb_off_t off, tdb_len_t len, bool convert);
495 /* Release result of tdb_access_read/write. */
496 void tdb_access_release(struct tdb_context *tdb, const void *p);
497 /* Commit result of tdb_acces_write. */
498 enum TDB_ERROR tdb_access_commit(struct tdb_context *tdb, void *p);
500 /* Convenience routine to get an offset. */
501 tdb_off_t tdb_read_off(struct tdb_context *tdb, tdb_off_t off);
503 /* Write an offset at an offset. */
504 enum TDB_ERROR tdb_write_off(struct tdb_context *tdb, tdb_off_t off,
507 /* Clear an ondisk area. */
508 enum TDB_ERROR zero_out(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len);
510 /* Return a non-zero offset between >= start < end in this array (or end). */
511 tdb_off_t tdb_find_nonzero_off(struct tdb_context *tdb,
516 /* Return a zero offset in this array, or num. */
517 tdb_off_t tdb_find_zero_off(struct tdb_context *tdb, tdb_off_t off,
520 /* Allocate and make a copy of some offset. */
521 void *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
523 /* Writes a converted copy of a record. */
524 enum TDB_ERROR tdb_write_convert(struct tdb_context *tdb, tdb_off_t off,
525 const void *rec, size_t len);
527 /* Reads record and converts it */
528 enum TDB_ERROR tdb_read_convert(struct tdb_context *tdb, tdb_off_t off,
529 void *rec, size_t len);
531 /* Bump the seqnum (caller checks for tdb->flags & TDB_SEQNUM) */
532 void tdb_inc_seqnum(struct tdb_context *tdb);
534 /* Adds a stat, if it's in range. */
535 void add_stat_(struct tdb_context *tdb, uint64_t *stat, size_t val);
536 #define add_stat(tdb, statname, val) \
538 if (unlikely((tdb)->stats)) \
539 add_stat_((tdb), &(tdb)->stats->statname, (val)); \
543 /* Lock/unlock a range of hashes. */
544 enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
545 tdb_off_t hash_lock, tdb_len_t hash_range,
546 int ltype, enum tdb_lock_flags waitflag);
547 enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
549 tdb_len_t hash_range, int ltype);
551 /* For closing the file. */
552 void tdb_unlock_all(struct tdb_context *tdb);
554 /* Lock/unlock a particular free bucket. */
555 enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
556 enum tdb_lock_flags waitflag);
557 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off);
559 /* Serialize transaction start. */
560 enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype);
561 void tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
563 /* Do we have any hash locks (ie. via tdb_chainlock) ? */
564 bool tdb_has_hash_locks(struct tdb_context *tdb);
566 /* Lock entire database. */
567 enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
568 enum tdb_lock_flags flags, bool upgradable);
569 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype);
570 enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb);
572 /* Serialize db open. */
573 enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb,
574 enum tdb_lock_flags flags);
575 void tdb_unlock_open(struct tdb_context *tdb);
576 bool tdb_has_open_lock(struct tdb_context *tdb);
578 /* Serialize db expand. */
579 enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype);
580 void tdb_unlock_expand(struct tdb_context *tdb, int ltype);
581 bool tdb_has_expansion_lock(struct tdb_context *tdb);
583 /* If it needs recovery, grab all the locks and do it. */
584 enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb);
587 enum TDB_ERROR tdb_transaction_recover(struct tdb_context *tdb);
588 tdb_bool_err tdb_needs_recovery(struct tdb_context *tdb);
591 enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
592 enum TDB_ERROR ecode,
593 enum tdb_log_level level,
594 const char *fmt, ...);
597 void tdb_trace(struct tdb_context *tdb, const char *op);
598 void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
599 void tdb_trace_open(struct tdb_context *tdb, const char *op,
600 unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
601 void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
602 void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
603 void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
605 void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
606 TDB_DATA rec, int ret);
607 void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
608 TDB_DATA rec, TDB_DATA ret);
609 void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
610 TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
612 void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
613 TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
615 #define tdb_trace(tdb, op)
616 #define tdb_trace_seqnum(tdb, seqnum, op)
617 #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
618 #define tdb_trace_ret(tdb, op, ret)
619 #define tdb_trace_retrec(tdb, op, ret)
620 #define tdb_trace_1rec(tdb, op, rec)
621 #define tdb_trace_1rec_ret(tdb, op, rec, ret)
622 #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
623 #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
624 #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
625 #endif /* !TDB_TRACE */